aoeusnth1

You say that as if Uber's playbook didn't work. Try this: https://www.google.com/finance/quote/UBER:NYSE

There is already evidence provided of it! METR time horizons is going up on an exponential trend. This is literally the most famous AI benchmark and already mentioned in this thread.

https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...

https://metr.org/blog/2025-07-14-how-does-time-horizon-vary-...

> Language model capability at generating text output.

How would you put this on a graph?

It has! CLs/engineer increased by 10% this year.

LLMs from late 2024 were nearly worthless as coding agents, so given they have quadrupled in capability since then (exponential growth, btw), it's not surprising to see a modestly positive impact on SWE work.

Also, I'm noticing you're not explaining yourself :)

We're very clearly seeing exponential progress - even above trend, on METR, whose slope keeps getting revised to a higher and higher estimate each time. Explain your perspective on the objective evidence against exponential progress?

Is there zero skill in managing agents?

Is that really what he's saying here?

He's not against the technology, I think he's just feeling like there's a lot of potential that he's not quite grasping yet.

This is also what I did. Actually, Claude did it.

It lets Claude directly type into Codex as if it were the user, or vice versa

It did come from Claude, though, not Anthropic.

You're right, sqrt(E[d^2]) > E[|d|] because of Jensen's inequality. The latter is about 0.8 sqrt(N).

Can you give an example of some ridiculous comments

tokens/s don't match, so unlikely

I think it's good, they're raising the size (and price) of flash a bit and trying to position Flash as an actually useful coding / reasoning model. There's always lite for people who want dirt cheap prices and don't care about quality at all.

In what way were the sentences content heavy? It's quite repetitive, and often the meaning of a section of it will be split into individual fragments.

I get it.

One sentence pragraphs feel punchy.

It feels like you're writing copy for an Apple ad.

..but it only works when it's in another medium, in a shorter format. In this form, it's just exhausting.

This is really cool! Are you considering publishing a paper on it? This seems conceptually similar to landmark MDS / Isomap, except using PCA on the landmark matrix instead of MDS. (https://cannoodt.dev/2019/11/lmds-landmark-multi-dimensional...)

Cool name, with both hints at "decid[e]" and the graphs.

I'd be interested in integrating this with bug systems of decisions / goals, with actions being comments on those bugs (for work purposes) instead of having a custom deciduous-only DB.

Is this meant to be open source? I don't see a LICENSE.

It's possible to use the Gemini "ask me about this screen" to OCR the selected area of the screenshot. I guess that might be more efficient in some contexts then trying to use the native text select.

Curious, where do you get your news? What gives you the impression that coal or gas would possibly be cheaper than solar?

Is this what he meant by making the company more masculine?

Maybe nothing, but can you afford to prove that in court?

Regressive taxes aren't bad inherently bad. Regressive spending is bad.

In this case, you have a regressive tax with a huge positive side effect due to taxing an externality. If the funds are also spread into progressive services it can be a net positive for all income brackets.

15x over 25 years is a reasonable ask for probably half of tech companies because you don't need 35%/y do accomplish it, you only need 11.4%.

Their math is just wrong. If they need to grow 15x over 25 years, they only need a modest 11.4% growth per year.

15^(1/35) = 1.11

1500^(1/35) = 1.34

Did the author maybe put a few extra zeros in their calculator when figuring out the annual revenue growth which would equal 15x over a 25 year span?

Lots of "it's not X. It's Y."

Bullet points I can forgive, it's a common blog post writing style. But the ranty prose here definitely has a whiff of silicon.

What's the history on the end to the loudness war? Do streaming services renormalize super compressed music to be quieter than the peaks of higher dynamic range music?

Token economics are very sound - it's the training which is expensive.

Tokens/week have gone up 23x year-over-year according to https://openrouter.ai/rankings. This is probably around $500M-1B in sales per year.

The real question is where the trajectory of this rocket ship is going. Will per-token pricing be a race to the bottom against budget chinese model providers? Will we see another 20x year year over the next 3 years, or will it level out sooner?

The reason why vertical integration is important for AI investment is that if AI is commoditized, then that AI-acceleration will costs pennies for drugs that are worth billions.

I don't see how OpenAI or Google can profit from drug discovery. It's nearly pure consumer surplus (where the drug companies and patients are the consumers).

The utility is that it's predictive of future observations, like all good language.

This looks related to (maybe descended from?) Nikita Lisitsa’s particle life project: https://lisyarus.github.io/blog/posts/particle-life-simulati...